Documentation · Reference genes
Choosing reference genes
A fold change is only as stable as the genes it is normalized to. Ampliova ranks candidate reference genes with four published measures, combines them into one ranking, and proposes the pair to normalize with.
Validation batches
Stability is evaluated in two places. The first is a dedicated batch type, Reference-gene validation, chosen when you create the batch. Declare every candidate as a reference gene and run them on samples that cover the conditions of the experiment you are preparing: a gene stable in controls and unstable under treatment is precisely the one to exclude. There are no target genes and no fold changes in such a batch.
The second is an RT² Profiler batch: before its analysis runs, the housekeeping genes of the panel are ranked the same way and the best pair is preselected. You confirm the genes to normalize with, or pick others, and the analysis runs on them.
Technical replicates are averaged first, and a sample enters the analysis only if it has a value for every candidate. Samples missing one are left out and listed. The measures run on Cq values, as RefFinder and RefSeeker do, which amounts to assuming the same efficiency for every candidate.
The four measures
All four are computed by the RefSeeker R package [15], which implements them as RefFinder does. For each, a lower value means a more stable gene.
geNorm M
For each candidate, the standard deviation across samples of its log2 ratio to each other candidate, averaged over the other candidates [10]. A gene that rises and falls with the others scores low. Ampliova flags M ≥ 0.5, the value usually given for homogeneous samples; 1.0 is the looser limit for heterogeneous ones [7].
BestKeeper
The spread of the gene’s own Cq values across samples, as the mean absolute deviation that BestKeeper reports under the name SD [12]. Pfaffl and colleagues considered a spread above 1 cycle inconsistent; Ampliova flags from 0.5, which is stricter.
ΔCq
The mean standard deviation of the Cq difference between the gene and each other candidate across samples [13]. A stable gene keeps a constant distance from the others.
NormFinder
A model-based estimate of each gene’s variation that separates the variation within groups from the variation between them [11], which makes it the measure most sensitive to a gene that shifts with the condition. The groups of the batch are used when there are at least two. There is no published cutoff; Ampliova flags values from 0.15.
Combined rank and best pair
The four rankings are combined as RefFinder does [14]: each gene’s combined score is the geometric mean of its four ranks, and the genes are ordered by it. It is a consensus, not a fifth measure.
Since at least two reference genes are needed [10, 2], what matters in the end is the pair. The proposed best pair is the pair with the lowest combined NormFinder stability, computed for every pair of candidates, with the groups of the batch when there are two or more. When NormFinder cannot score pairs, the two genes with the best combined rank are proposed instead. The pair is only a proposal: on an RT² batch you confirm it before the analysis runs, and after a validation batch it tells you which genes to declare in the experiments that follow.
The geNorm pairwise variation Vn/n+1 is also reported, adding genes in the order of the combined rank. A value below 0.15 suggests that n genes are enough and that an (n + 1)th adds little [10].
Reading the results
| Gene | geNorm M | BestKeeper | ΔCq | NormFinder | Combined rank |
|---|---|---|---|---|---|
| RPLP0 | 0.34 | 0.28 | 0.41 | 0.07 | 1 |
| HPRT1 | 0.39 | 0.36 | 0.46 | 0.10 | 2 |
| GAPDH | 0.62 | 0.71 | 0.75 | 0.22 | 3, flagged |
An illustrative result. RPLP0 and HPRT1 agree across all four measures and would form the pair; GAPDH exceeds every flag. When all four measures agree, the choice is easy. When they do not, NormFinder deserves the most weight if your groups differ, since it is the only one that looks at the groups, and geNorm the least, since it rewards genes that co-vary, even when they co-vary because both respond to the treatment.
Several plates
Plate-to-plate variation reads as instability. Run all candidates of a sample on the same plate, or turn on inter-run calibration with a calibrator for every candidate on every plate. See several plates.